Hi All,

Chris and I have just uploaded the latest revisions of our BOINC and SETI clients for Linux:

http://www.pperry.f2s.com/downloads.htm

There are 2 new versions of the BOINC core client based on v4.19 for i686 and Athlon XP. These are our 3rd revision and are numbered v4.19.3 accordingly. The level of benchmark (Whetstone/Dhrystone) performance is now virtually identical with both the i686 and Athlon XP clients on an Athlon XP processor. Therefore, as the i686 client is now so highly optimized we have decided to discontinue BOINC clients for other specific CPU architectures unless we find there is a genuine performance increase. The i686 client gives about +20% and the Athlon XP client about +13% over our previous BOINC clients in combined Whetstone/Dhrystone benchmarks (ie requested credits).

There are also 2 new versions of the SETI client based on v4.07 for i686 and Athlon XP. Again, these are our 3rd revision and are numbered v4.07.3 accordingly. Benchmarks with the Berkeley reference test work unit on an Athlon XP processor show the i686 client to give about +41% and the Athlon XP client to be about +59%. These are significantly faster than our previous revision 2 clients which were about +30%.

We have thoroughly tested the Athlon XP client and results validate fine.

The i686 is a truely generic client and should run on any processor architecture but we have not tested the i686 client on other processor architectures to see how well it will perform. We would appreciate feedback from anyone who runs this client on other processors. This client should be ideal for users with older (non-XP) Athlon processors and we would expect about +40% levels of performance. Of course this client will also run on Intel processors, but users may find Metod's excellent Intel clients to be faster (hopefully Metod will benchmark this client for comparison).

New revisions of our 64-bit client for AMD64 (x86-64) for both BOINC and SETI will hopefully be available shortly.

WARNING: Make sure you upload and report any outstanding work before installing the revised SETI clients to avoid losing any work.

Enjoy

Ned

---

*** My Guide to Compiling Optimised BOINC and SETI Clients ***
*** Download Optimised BOINC and SETI Clients for Linux Here ***

WARNING: Make sure you upload and report any outstanding work before installing the revised SETI clients to avoid losing any work.

or use a speziall app_info.xml, like this

http://www.andyk.de/app_info.xml

with this you loose no Work

Greetings from Germany NRW
Ulli S@h Berkeley's Staff Friends Club m7 ©

Hi Ned,

Just installed on a Duron 900. Will report back when I have some validated results :-)

Thanks for all the hard work from you, and your team of optimisers

Dunc

---

Is there such a thing as a Windows optimised client for amd or intel?

Live long and crunch!

Paul.

<blockquote>Is there such a thing as a Windows optimised client for amd or intel?

Live long and crunch!

Paul.</blockquote>

Yes. see threads:


Compiling faster Windows client with Intel C++ compiler (and fftw?)

2nd: Compiling faster Windows client with Intel C++ compiler (and fftw?)

3nd: Compiling faster Windows client with Intel C++ compiler (and fftw?)

---

Ned,

I just tested your new version of the BOINC core client boinc_4.19.3_i686-pc-linux-gnu on an athlon thunderbird box. It didn't run at all.

Here is the output:

user@host:~/test/boinc$ ./boinc_4.19.3_i686-pc-linux-gnu
set_thread_area failed when setting up thread-local storage
Speicherzugriffsfehler


The last one need to be translated as it is german: segmentation fault
Any idea?
I'm just in the process of testing it on an intel box. I will post the results asap.

edit
It produces the same error on an intel P3 (Katmai). I was a bit surprised about the small size of your binaries, much smaller then the predecessor.
/edit

However, thanks again for your amazing work.

---

Hi Rattelschneck

Could you give me some details on what system you use?
Hardware as well as the OS, Distribution and Kernel version you are running.

The client runs here on Athlon-XP as well as on one P3 without any problems.
I use Fedora Core 3 on both machines. They currently use the Kernel 2.6.11.

regards
Chris

---

Chris Bosshard
Visit my homepage
astroinfo SETI page

<blockquote>
Could you give me some details on what system you use?
Hardware as well as the OS, Distribution and Kernel version you are running.
</blockquote>

Hi Chris,

here are the details as requested.

1. Athlon Thunderbird
athlon thunderbird 900 mhz
Debian GNU/Linux 3.1 (sarge)
kernel 2.4.27-2-k7

2. Pentium III
pentium 3 450 mhz
Debian GNU/Linux 3.1 (sarge)
kernel 2.4.27-2-686

Hope that helps.
And many thanks from me for the excellent work you and Ned did so far!

regards
rattelschneck

Not to look a gift horse in the mouth or anything... but these XP clients are *so* fast, my XP2200 and XP2400 are theatening my Athlon64 3200+ !!

An equally souped-up AMD64 client would be way cool! :)

---

Information update:

I just had a try on my Athlon-XP box (Athlon XP 2400+), using your boinc_4.19.3_athlon-xp-pc-linux-gnu client. I'm really sorry, but it won't run on that either. I got the same messages again.

knoppix@1[test]$ ./boinc_4.19.3_athlon-xp-pc-linux-gnu
set_thread_area failed when setting up thread-local storage
Speicherzugriffsfehler



The details:

Athlon-XP
OS = KNOPPIX 3.7
kernel 2.4.27

cpuinfo:
knoppix@1[test]$ cat /proc/cpuinfo
processor : 0
vendor_id : AuthenticAMD
cpu family : 6
model : 8
model name : AMD Athlon(tm) XP 2400+
stepping : 1
cpu MHz : 2019.944
cache size : 256 KB
fdiv_bug : no
hlt_bug : no
f00f_bug : no
coma_bug : no
fpu : yes
fpu_exception : yes
cpuid level : 1
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 mmx fxsr sse syscall mmxext 3dnowext 3dnow
bogomips : 4023.91

KNOPPIX is based on Debian, AFAIK.
Just found a harddisk with Gentoo Linux stage 1, optimized for Athlon-XP, which was lying around. I will do a quick test with this one for now and report the result later on.

regards
rattelschneck

rattelschneck,

Thanks for the feedback, we'll look into it. I'd only tested the boinc clients on an FC2 Athlon XP box on which they ran fine. Sorry for any inconvenience.

In the meantime, do you still have a copy of my older boinc clients to use?

I'll try an put the last 4.19 versions back on my site in case anyone else has problems.

@ Abject - we hope to have a new AMD64 (x86-64) version soon. In the meantime, the new Athlon XP (v4.07.3) version may actually be faster than the old AMD64 version (v4.07.2).

Ned

---

*** My Guide to Compiling Optimised BOINC and SETI Clients ***
*** Download Optimised BOINC and SETI Clients for Linux Here ***

<blockquote>
Thanks for the feedback, we'll look into it. I'd only tested the boinc clients on an FC2 Athlon XP box on which they ran fine. Sorry for any inconvenience.
</blockquote>

Never mind, no inconvenience at all. It's a pleasure for me doing some testing if it helps to sort out some issues.

<blockquote>
In the meantime, do you still have a copy of my older boinc clients to use?
</blockquote>

Sure, still using your older clients on most of my boxes.

Runnig your new client on Gentoo Linux wasn't successful, too. Unfortunately. The same error messages.

rattelschneck

Chris and Ned,

That XP is a "Thoroughbred" core --- vendor_id : AuthenticAMD...cpu family : 6...model : 8.

Other XP cores are Palomino, Morgan and Barton.

Really not sure what the problem is here. I tested both clients on a Athlon XP (palomino) and they tested fine. Just also tested on another older mobile Athlon XP running FC1 and a 2.4 series kernel, and again they both run fine.

The athlon xp client uses -march=athlon-xp so should run on _any_ Athlon XP (Palomino, thoroughbred, barton). It won't however run on older athlons (non-XP, thunderbird), but you were using the i686 client on that.

Is anyone else able to reproduce this error??

Ned

---

*** My Guide to Compiling Optimised BOINC and SETI Clients ***
*** Download Optimised BOINC and SETI Clients for Linux Here ***

<blockquote>
Is anyone else able to reproduce this error??
</blockquote>

Nope. All 3 of your new clients run fine here.

AMD64 - both x86-64 and XP clients.
XP - XP client
Athlon - i686 client
Duron - i686 client

All under Gentoo GNU/Linux.

Thank you for all your hard work. :)

---

Be lucky

Neil

OK, since it seems only me having this problem, i just downloaded Fedora Core 3 rescuecd. Fortunately the rescue cd contains enough stuff to do some _very_ basic testing.
So i downloaded the boinc_4.19.3_i686-pc-linux-gnu client from Ned and Chris and started it on the Athlon Thunderbird box from my very first test.
Big surprise here - it works!

Same hardware, another linux distro.
Just considering if I should stay with the rescuecd for one night or maybe two for do some further testing.

update
Just recognized that the rescuecd is using 2.6 kernel. Somebody else running kernel 2.4?
/update

<blockquote>
update
Just recognized that the rescuecd is using 2.6 kernel. Somebody else running kernel 2.4?
/update


</blockquote>

I tested on FC1 running 2.4 kernel:

2.4.22-1.2199.4.legacy.nptl #1 Sun Feb 20 18:49:30 EST 2005 i686 athlon i386 GNU/Linux

and it was fine, so I don't think it's a 2.4 kernel issue. Eric Korpela did warn me that static clients aren't always TOTALLY static. Sounds like it's maybe just a glitch on your particular distro??

Re the client size question - Chris stripped the latest boinc clients whereas I hadn't stripped any of my previous clients. When I stripped an older boinc client, the size was similar. I wouldn't think that this would cause this problem though.

Ned

---

*** My Guide to Compiling Optimised BOINC and SETI Clients ***
*** Download Optimised BOINC and SETI Clients for Linux Here ***

<blockquote>
and it was fine, so I don't think it's a 2.4 kernel issue. Eric Korpela did warn me that static clients aren't always TOTALLY static. Sounds like it's maybe just a glitch on your particular distro??
</blockquote>

Ned,

I've just installed kernel 2.6.8-2-686 on one of my Debian boxes.

And now your boinc_4.19.3_i686-pc-linux-gnu client works great along with your setiathome_4.07.3-i686-fftw3-static-pc-linux-gnu client.

FYI:
Pentium II 450 mhz XEON, (512k cache), Debian GNU/Linux 3.1 (sarge) kernel 2.6.8-2-686
Benchmark results:
472 double precision MIPS (Whetstone)
972 integer MIPS (Dhrystone)

regards
rattelschneck

Great stuff :)

So it would appear it was a 2.4 kernel issue after all?

Strange

Please let us know how the i686 seti client compares to other clients you've run.

Ned

---

*** My Guide to Compiling Optimised BOINC and SETI Clients ***
*** Download Optimised BOINC and SETI Clients for Linux Here ***

<blockquote>Not to look a gift horse in the mouth or anything... but these XP clients are *so* fast, my XP2200 and XP2400 are theatening my Athlon64 3200+ !!

An equally souped-up AMD64 client would be way cool! :)</blockquote>

A new 64-bit BOINC client (v4.19.3) is now available :)

We're having a few problems with the new version (4.07.3) of the 64-bit SETI client - it's actually slower than the current version (4.07.2), and we're not seeing the same levels of performance increases with our latest tweaks that we saw for the 32-bit clients.

Chris is currently benchtesting the new 32-bit Athlon-XP client (v4.07.3) on his FC3 64-bit system as we suspect this may now be the fastest client currently available for AMD64 users (faster than our current v4.07.2 AMD64 client and Metod's P4 Intel client).

Here are Chris's current benchmark results:

Reference 4.02 Berkeley: 9147s
Our v4.07.2 x86_64: 6466s +41,5%
Aki's v4.7 x86_64: 6975s +31%
Metod's 4.07 ICC-P4: 6530s +40%
Our latest setiathome_4.07.3c_x86_64: 8048s +13% ???

As you can see, somethings definately not quite right :(

The new Athlon-XP client (v4.07.3) gives about +59% on an Athlon-XP, but we don't know how that performance will translate onto an AMD64 system yet (I'd estimate somewhere around +50% probably making it the fastest currently available option)

Ned

---

*** My Guide to Compiling Optimised BOINC and SETI Clients ***
*** Download Optimised BOINC and SETI Clients for Linux Here ***